Table of Contents  

Click to Review
"Preoperative prediction model of outcome after cholerystectomy for symptomatic gallstones" by Borly et.al, Scandinavian Journal of Gastroenterology, 1999, Vol. 34, pages 1144-1152. Reprinted from Scandinavian Journal of Gastroenterology by permission of Taylor & Francis. For further information on the journal, please visit the journal's home page at www.sjgweb.net.

Click here to get PDF Viewer

INSTRUCTIONS: For optimal educational effect, we suggest reading this paper first and answering the questions for yourself, then comparing your response with our answers. However, questions also serve as an illustration to a topic of personal interest, for example "missing values" or "validation."

The paper is an interesting example of studying the predictive value of pain and symptoms for outcome after surgery. Unfortunately, the statistical analysis has a number of weaknesses, as partly acknowledged by the authors, and does not comply to the principles of prognostic modeling as outlined in the chapter. To guide you through this paper, we consider the 7 steps in the development of a prognostic model:

Click on any question below to see the answer.

1. Preliminary Steps

a.	What was the aim of the analysis?
b.	Is this aim specific enough, and in line with the methodological approach, especially the development of prediction model?
c.	Do you think the data set is adequate to develop a prediction model?
d.	How was an impression of the data under study obtained?
e.	How were missing values treated for the logistic regression analysis?
f.	Why were missing values not omitted from the analysis?
g.	Is it correct that the imputation technique causes no bias?

2. Coding of covariables

In this study, we can identify three types of covariables: dichotomous, categorical, and continuous.

a.	Dyspeptic symptoms were combined in a ranking scale ranging from 0 to 16. How was the scale considered in the logistic regression analysis?

3. Selection of covariables

a.	How were covariables selected for inclusion in the multivariable models?
b.	How many covariables were considered in the analysis?

4. Estimation of regression coefficients

a.	How were the regression coefficients estimated?

5. Evaluation of performance

a.	Which performance measures were used in the paper?

6. Validation

a.	What type of validation was performed?
b.	Do you agree with the statement in the Discussion (p 1151, first column) that the number of variables tested was <10% of the total number of patients, and that this is in concordance with commonly accepted rules?
c.	What are the expected results of internal validation of the predictive model?

7. Presentation of model

a.	How was the final model presented?
b.	Is the model suitable for application in clinical practice?

 

In conclusion, this study is a nice illustration of predictive research with pain and symptom variables. Many of the issues related to prognostic modeling could be addressed, as discussed in the chapter. We conclude that the development of a prediction model on such a small data set should not have been attempted. The study may have some value as an exploratory analysis of potential predictors that should be evaluated in future studies.